Welcome to Yan Xia's Page ！
Welcome to my website!
I would like to tackle climate change with ML in my future studies.
Hybrid model. Build a hybrid model by combining physical-processed models and data-driven ML to represent subgrid processes (e.g., cloud microphysics and convection); predict extreme events (e.g., extreme precipitation); and improve understanding of aerosol-cloud-climate interactions.
Data-constrained hybrid model. Tackle earth monitoring data (e.g., remote sensing) with ML by retrieving and inverting vital biophysical parameters to correct the established hybrid model.
Uncertainty estimation and model interpretability. Develop theories and methods (e.g., Bayesian/probabilistic inference) to better understand the established hybrid model and fundamental scientific questions.
Physics-informed ML(NN). Embed Physics-informed ML(NN) in all above processes by adding physical constraints in the loss function or modifying the architecture of neural networks to improve model accuracy and deepen the development of ML.